Biomedical records contain a vast amount of unstructured information that can provide valuable insights for patient care and clinical research. These records capture important details about diagnoses, treatments, outcomes, and clinical decision-making, but their unstructured nature makes them difficult to analyze at scale. To overcome these limitations, researchers have turned to natural language processing (NLP) techniques to automate biomedical knowledge extraction, the process of identifying and structuring relevant information from unstructured biomedical text, and knowledge retrieval, the process of identifying and accessing relevant information from existing biomedical resources. Despite substantial progress, existing systems remain constrained by performance gaps, limited generalizability, and restricted applicability across diverse domains and conditions, which hinders their reliable deployment in real-world settings.  This dissertation aims to advance the performance, robustness, and practical applicability of methods for identifying, grounding, retrieving, and structuring biomedical information. The broader goal is to transform complex biomedical text into structured and searchable knowledge that can better support clinical analysis, knowledge discovery and biomedical research.
Specifically, this work develops methods to: (1) structure clinical data to facilitate knowledge discovery and downstream analysis, (2) improve the reliability of biomedical search by surfacing the most accurate and relevant matches, and (3) expand retrieval coverage to identify a broader set of relevant results in large-scale biomedical databases. Together, these contributions address complementary challenges in biomedical NLP and aim to make clinical and scientific information more accessible, reliable and useful for research and decision support.